Automatic Building Roof Plane Extraction in Urban Environments for 3D City Modelling Using Remote Sensing Data
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe article is generally well structured and well written.
More detailed comments are in the atached PDF.
Comments for author File: Comments.pdf
Author Response
We would like to express our gratitude to the reviewer for the evaluation of our work and for the time spent and effort to provide us with helpful comments. We have addressed all indicated suggestions and recommendations. Please see the attached file. We have provided a new version of the paper with the changes according to the reviewer's suggestions. Please note that the page numbers in this response correspond to the file which is showing the track changes (all markup). We also provide a file without track changes where the figures and empty spaces are arranged.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors The paper titled "Automatic Building Roof Plane Extraction in Urban Environments for 3D City Modelling Using Remote Sensing Data" introduces a framework based on H.E.A.T. to automatically extract building inner roof planes and then use them to generate a 3D model at LOD2. The research work looks promising and can be accepted if the following questions can be answered: 1. What factors contributed to the superior performance of the model trained on a combined dataset for the Sofia dataset, as opposed to its performance on the Oude Markt dataset? How to improve performance on Sofia dataset? 2. How does using cadastral information in the framework improve the accuracy and efficiency of building roof plane extraction? 3. Why was H.E.A.T. chosen as a desired model for extracting roof plane structures from aerial imagery? 4. How does the framework's performance vary across different roof structure topologies? 5. What is the scale of training data required to impact the model's ability to delineate obscured edges? 6. What are the limitations of the proposed framework, and how can it be further improved to address issues like topological inconsistencies and incomplete corner prediction? 7. Can the framework be adapted to handle more complex and dense roof topologies without requiring additional post-processing? 8. Why were building footprints not automatically extracted from aerial imagery through deep learning methods?Minor: 1. The article contains numerous spelling errors and ambiguous lines, such as those found in line numbers 90, 99, 172, 174, etc. as well as in Figure 5.
Author Response
We would like to express our gratitude to the reviewer for the evaluation of our work and for the time spent and effort to provide us with helpful comments. We have addressed all indicated suggestions and recommendations. Please see the attached file. We have provided a new version of the paper with the changes according to the reviewer's suggestions. Please note that the page numbers in this response correspond to the file which is showing the track changes (all markup). We also provide a file without track changes where the figures and empty spaces are arranged.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis is a very well written paper that deals with an interesting and current problem. From the user's perspective, it becomes clear what results can be expected when reconstructing roofs using deep learning. The transformer network HEAT is used (out of the box) to reconstruct a planar graph of building roof edges from aerial images in vector format. Then an LoD2 model is derives from a digital elevation model. Results for three small test areas are discussed. The paper does not really describe any new methods, but current methods are used and analyzed in a test context.
Do you think that modelling from ALS data could be improved by additionally applying HEAT to aerial images?
line 211: 3D city modelling methodology: Could you shortly describe (without having to look into [27]) how ridge lines (intersections of planes) and step edges are handled in this process? Is there an additional alignment?
line 259: add field to the table? This is not clear and a database is not mentioned.
line 355: Is there an explanation for the value 0.7? It appears to be quite small?
line 406: is it number of buildings or number of planes? The range 0-5 m is quite wide. How is the distribution within this range?
In addition to the HEAT network, there are alternatives, e.g. PPGNet:
Simon Hensel, Steffen Goebbels and Martin Kada: Building roof vectorization with PPGNet. ISPRS Archives XLVI-4/W4-2021, S. 85–905, 2021, https://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XLVI-4-W4-2021/85/2021/
Comments on the Quality of English Languagetypos:
line 90: Therefoe -> Therefore
line 99: otlines -> outlines
line 209: it -> is
Author Response
We would like to express our gratitude to the reviewer for the evaluation of our work and for the time spent and effort to provide us with helpful comments. We have addressed all indicated suggestions and recommendations. Please see the attached file. We have provided a new version of the paper with the changes according to the reviewer's suggestions. Please note that the page numbers in this response correspond to the file which is showing the track changes (all markup). We also provide a file without track changes where the figures and empty spaces are arranged.
Author Response File: Author Response.pdf